An ocean economy fish resource evaluation platform based on multi-source data fusion

Abstract

The application discloses a marine economic fish resource evaluation platform based on multi-source data fusion. The platform sets a target water area layer and multi-period monitoring, uses a sonar detection device to collect data to generate first fish school characteristics, collects images through an underwater camera device and generates second fish school characteristics through a preset image recognition model; a fish school characteristic matrix is constructed based on time window analysis, noise time periods are screened by using matrix similarity discrimination; the recognition process of the noise image set is dynamically adjusted in edge segmentation parameters, and the resource evaluation of economic fish is carried out by fusing and correcting the recognition result. Through multi-source data fusion analysis, difference comparison of fish school characteristics and dynamic image parameter adjustment, the application effectively improves the comprehensiveness, reliability and accuracy of fish school characteristic evaluation in complex water environment, can provide real-time dynamic data support for fishery resource management and marine ecological protection, and has a wide application prospect.

Description

Technical Field

[0001] This invention relates to the field of marine resource analysis, and more specifically, to a marine economic fish resource assessment platform based on multi-source data fusion. Background Technology

[0002] Currently, the assessment of marine economic fish resources mainly relies on single monitoring technologies, commonly including sonar detection, underwater image recognition, and fisheries survey statistics. However, due to the influence of marine environmental factors (such as water temperature stratification, salinity changes, water turbidity, and light intensity variations), the accuracy of traditional underwater image recognition often fluctuates significantly. For image segmentation, edge detection, and classification of target economic fish species, there are often large recognition errors. This is generally due to the complexity of the underwater environment, fluctuations in light intensity, and the presence of obstacles, making image noise points complex and difficult to predict. Furthermore, existing technologies lack corresponding dynamic adjustment methods for edge detection, resulting in weak image segmentation capabilities and difficulty in controlling recognition accuracy. Using sonar alone for fish school detection often only detects data such as fish density and range, failing to efficiently and quickly detect and identify fish species. Therefore, for fish resource identification and classification using underwater images, there is an urgent need for technologies that integrate multi-source data for dynamic analysis and assessment, thereby improving resource assessment capabilities in complex water conditions. Summary of the Invention

[0003] This invention overcomes the shortcomings of existing technologies and proposes a marine economic fish resource assessment platform based on multi-source data fusion.

[0004] The first aspect of this invention provides a method for assessing marine economic fish resources based on multi-source data fusion, comprising:

[0005] S1: Within the target sea area, set the target water layer and collect data within a preset monitoring period. Collect sonar detection data based on the sonar detection device, and evaluate the fish school characteristics of economic fish based on the sound wave characteristic feedback to generate the first fish school characteristics for multiple time periods.

[0006] S2: The underwater camera device is used to collect images of the target water layer to obtain an image set. Edge segmentation parameters are set in the preset image recognition model. The image set is segmented, identified and evaluated for various economic fish species, and fish school characteristics are generated for multiple time periods.

[0007] S3: Set time windows based on preset monitoring cycles, combine the characteristics of the first fish group and the characteristics of the second fish group to generate the first feature matrix and the second feature matrix for each time window, use spectral decomposition to determine the similarity of the feature matrices within each time window, and filter out the noise time periods based on the similarity.

[0008] S4: Select a set of noisy images from the image set based on the noise time period, and dynamically adjust the edge segmentation parameters according to the similarity to perform secondary fish identification, and dynamically correct the fish identification results.

[0009] In this solution, S1 specifically refers to:

[0010] Monitoring devices are set up according to the target sea area, and water monitoring is carried out on the target water layer within a preset monitoring period;

[0011] Based on the sonar detection data collected by the sonar detection device, the sonar detection data is processed by adaptive median filtering to obtain time-domain sonar data;

[0012] The time-domain sonar data is converted into sonar feature information, including frequency domain data, by Fourier transform, and the amplitude, frequency, phase and sonar intensity features of the sound wave are extracted.

[0013] By combining the fish sonar feature database, the sonar feature information is used to identify economically important fish species and evaluate fish characteristics to obtain information on fish school size, fish school density, and fish school quantity.

[0014] Based on the monitoring frequency, the preset monitoring period is divided into multiple time periods. For each time period, the corresponding fish school range size, fish school density, and fish school number information are analyzed as the first fish school feature in multiple dimensions.

[0015] In this solution, S2 specifically refers to:

[0016] Based on an underwater camera device, images are acquired at a preset monitoring cycle for the target water layer, and an image set is obtained;

[0017] The image acquisition frequency is kept consistent with the sonar detection and analysis frequency;

[0018] The image set is preprocessed by grayscale conversion, enhancement, and noise reduction;

[0019] In the preset image recognition model, the Sobel edge detection operator is used to perform image edge detection and image segmentation, and the default edge segmentation parameters are set.

[0020] Edge segmentation parameters include the kernel size and gradient magnitude threshold;

[0021] The Sobel edge detection operator is used to detect edges in the image set and define edge points to form edge regions.

[0022] Fish targets are segmented and classified based on edge regions to obtain fish identification results;

[0023] Based on the fish identification results, the fish school feature information of multiple time periods is statistically analyzed in the image, and a second fish school feature is generated.

[0024] In this solution, S3 specifically refers to:

[0025] A time window is set based on the preset monitoring cycle and the preset interval number N;

[0026] A time window includes multiple time periods;

[0027] Within a time window, the first feature matrix under that time window is constructed using time points as one-dimensional information and the characteristics of the first fish group as two-dimensional information.

[0028] The second feature matrix for this time window is constructed based on the features of the second fish swarm.

[0029] In this solution, S3 further includes:

[0030] Introduce a time window as the analysis period;

[0031] The first and second eigenma matrices are decomposed using spectral decomposition, resulting in multiple eigenvectors and corresponding eigenvalues. These eigenvectors are labeled as the first group and the second group based on their respective eigenma matrices.

[0032] We introduce cosine distance to calculate the similarity between two sets of feature vectors. Specifically, we select a feature vector from the first set of feature vectors and calculate the average distance D between it and the second set of feature vectors.

[0033] Calculate the average distance D between all feature vectors in the first group and the second group, and average all D values ​​as the difference between the two groups of vectors.

[0034] The time window is moved cyclically, with a step size of N1 time periods, and the difference between the two sets of vectors corresponding to each time window is calculated.

[0035] The time windows with differences exceeding a preset threshold are marked, and the corresponding time periods are designated as noise time periods.

[0036] In this solution, S4 specifically refers to:

[0037] Based on the noise time period, a set of noisy images is selected from the image set;

[0038] The edge segmentation parameters are dynamically set according to the difference value corresponding to the time window, and secondary fish recognition is performed based on the preset image recognition model. The recognition results are updated in real time, and the corrected fish recognition results are sent to the preset terminal.

[0039] In this solution, the preset terminal includes a computing terminal and a mobile terminal.

[0040] In this solution, both the preset monitoring period and the time window include multiple time periods.

[0041] A second aspect of the present invention also provides a marine economic fish resource assessment platform based on multi-source data fusion. The platform includes a memory and a processor. The memory includes a marine economic fish resource assessment program based on multi-source data fusion. When executed by the processor, the marine economic fish resource assessment program based on multi-source data fusion performs the following steps:

[0042] S1: Within the target sea area, set the target water layer and collect data within a preset monitoring period. Collect sonar detection data based on the sonar detection device, and evaluate the fish school characteristics of economic fish based on the sound wave characteristic feedback to generate the first fish school characteristics for multiple time periods.

[0043] S2: The underwater camera device is used to collect images of the target water layer to obtain an image set. Edge segmentation parameters are set in the preset image recognition model. The image set is segmented, identified and evaluated for various economic fish species, and fish school characteristics are generated for multiple time periods.

[0044] S3: Set time windows based on preset monitoring cycles, combine the characteristics of the first fish group and the characteristics of the second fish group to generate the first feature matrix and the second feature matrix for each time window, use spectral decomposition to determine the similarity of the feature matrices within each time window, and filter out the noise time periods based on the similarity.

[0045] S4: Select a set of noisy images from the image set based on the noise time period, and dynamically adjust the edge segmentation parameters according to the similarity to perform secondary fish identification, and dynamically correct the fish identification results.

[0046] A third aspect of the present invention also provides a computer-readable storage medium comprising a marine economic fish resource assessment program based on multi-source data fusion, wherein when the marine economic fish resource assessment program based on multi-source data fusion is executed by a processor, it implements the steps of the marine economic fish resource assessment method based on multi-source data fusion as described in any of the preceding claims.

[0047] This invention discloses a marine economic fish resource assessment platform based on multi-source data fusion. The platform sets target water layers and monitors multiple time periods. It uses sonar to collect data and generate first fish school features, and underwater cameras to acquire images and generate second fish school features using a pre-set image recognition model. A fish school feature matrix is ​​constructed based on time window analysis, and matrix similarity is used to filter noisy time periods. The edge segmentation parameters are dynamically adjusted during the recognition process of noisy image sets, and the fused and corrected recognition results are used for resource assessment of economic fish species. This invention, through multi-source data fusion analysis, comparison of differences in fish school features, and dynamic image parameter adjustment, effectively improves the comprehensiveness, reliability, and accuracy of fish school feature assessment in complex water environments. It can provide real-time dynamic data support for fisheries resource management and marine ecological protection, and has broad application prospects. Attached Figure Description

[0048] Figure 1 A flowchart of a method for assessing marine economic fish resources based on multi-source data fusion according to the present invention is shown;

[0049] Figure 2 The flowchart of the first fish swarm feature acquisition process of the present invention is shown;

[0050] Figure 3 The diagram shows a block diagram of a marine economic fish resource assessment platform based on multi-source data fusion according to the present invention. Detailed Implementation

[0051] To better understand the above-mentioned objects, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It is understood that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0052] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0053] Figure 1 A flowchart of a method for assessing marine economic fish resources based on multi-source data fusion according to the present invention is shown.

[0054] like Figure 1 As shown, the first aspect of this invention provides a method for assessing marine economic fish resources based on multi-source data fusion, comprising:

[0055] S1: Within the target sea area, set the target water layer and collect data within a preset monitoring period. Collect sonar detection data based on the sonar detection device, and evaluate the fish school characteristics of economic fish based on the sound wave characteristic feedback to generate the first fish school characteristics for multiple time periods.

[0056] S2: The underwater camera device is used to collect images of the target water layer to obtain an image set. Edge segmentation parameters are set in the preset image recognition model. The image set is segmented, identified and evaluated for various economic fish species, and fish school characteristics are generated for multiple time periods.

[0057] S3: Set time windows based on preset monitoring cycles, combine the characteristics of the first fish group and the characteristics of the second fish group to generate the first feature matrix and the second feature matrix for each time window, use spectral decomposition to determine the similarity of the feature matrices within each time window, and filter out the noise time periods based on the similarity.

[0058] S4: Select a set of noisy images from the image set based on the noise time period, and dynamically adjust the edge segmentation parameters according to the similarity to perform secondary fish identification, and dynamically correct the fish identification results.

[0059] Figure 2 The flowchart for obtaining the first fish swarm feature of the present invention is shown.

[0060] According to an embodiment of the present invention, S1 specifically includes:

[0061] Monitoring devices are set up according to the target sea area, and water monitoring is carried out on the target water layer within a preset monitoring period;

[0062] Based on the sonar detection data collected by the sonar detection device, the sonar detection data is processed by adaptive median filtering to obtain time-domain sonar data;

[0063] The time-domain sonar data is converted into sonar feature information, including frequency domain data, by Fourier transform, and the amplitude, frequency, phase and sonar intensity features of the sound wave are extracted.

[0064] By combining the fish sonar feature database, the sonar feature information is used to identify economically important fish species and evaluate fish characteristics to obtain information on fish school size, fish school density, and fish school quantity.

[0065] Based on the monitoring frequency, the preset monitoring period is divided into multiple time periods. For each time period, the corresponding fish school range size, fish school density, and fish school number information are analyzed as the first fish school feature in multiple dimensions.

[0066] This means the monitoring equipment includes sonar detection devices and underwater cameras, with specific settings determined by selecting the optimal monitoring location based on the target sea area. Multiple depth layers can be set for the target water area, for example, multiple layers can be analyzed, each 5-10m deep, with no overlap between adjacent layers. The preset monitoring period is the first relatively long monitoring duration, generally divided into multiple time periods for multi-time period fish school characteristic analysis. This fish school characteristic analysis includes both sonar-based analysis and image analysis. The detection sound wave frequency is 10-200kHz, the detection beamwidth is 1-5°, the detection distance can be set to 10-500m, and the sampling rate can be set to 150-1200Hz. Median filtering is used for filtering; other filtering methods can be selected based on the characteristics of sound wave noise. Image acquisition parameters can be set to a resolution of 1920×1080 pixels, a frame rate of 25fps, etc.

[0067] According to an embodiment of the present invention, step S2 specifically includes:

[0068] Based on an underwater camera device, images are acquired at a preset monitoring cycle for the target water layer, and an image set is obtained;

[0069] The image acquisition frequency is kept consistent with the sonar detection and analysis frequency;

[0070] The image set is preprocessed by grayscale conversion, enhancement, and noise reduction;

[0071] In the preset image recognition model, the Sobel edge detection operator is used to perform image edge detection and image segmentation, and the default edge segmentation parameters are set.

[0072] Edge segmentation parameters include the kernel size and gradient magnitude threshold;

[0073] The Sobel edge detection operator is used to detect edges in the image set and define edge points to form edge regions.

[0074] Fish targets are segmented and classified based on edge regions to obtain fish identification results;

[0075] Based on the fish identification results, the fish school feature information of multiple time periods is statistically analyzed in the image, and a second fish school feature is generated.

[0076] It can be understood here that keeping the image acquisition frequency consistent with the sonar detection and analysis frequency ensures that the frequency of image recognition results is consistent with the frequency of sonar recognition results, that is, a consistent number of fish school feature analyses and recognitions within each time period. The preset image recognition model can be an image recognition model such as YOLOv5, trained and classified by analyzing an existing fish feature database. Fish recognition results include information such as fish school markers, fish species, fish quantity, and fish school range in the image. Fish school feature information includes fish school size, fish school density, and fish school quantity information. The convolution kernel size and gradient magnitude threshold can be set to 3×3 and 100 (range 0-255).

[0077] According to an embodiment of the present invention, step S3 specifically includes:

[0078] A time window is set based on the preset monitoring cycle and the preset interval number N;

[0079] A time window includes multiple time periods;

[0080] Within a time window, the first feature matrix under that time window is constructed using time points as one-dimensional information and the characteristics of the first fish group as two-dimensional information.

[0081] The second feature matrix for this time window is constructed based on the features of the second fish swarm.

[0082] It can be understood here that the time window is a relatively short time length used to analyze fish school characteristics across multiple time periods. The time window is equal to a preset time period ÷ N. Here, each time period includes corresponding first and second fish school characteristics.

[0083] The feature matrix is ​​in the following form:

[0084]

[0085] V1-V6 represent relevant evaluation values, with a time window length of three time periods. The table above provides example data; specific feature analysis will be conducted based on actual evaluation values.

[0086] According to an embodiment of the present invention, step S3 further includes:

[0087] Introduce a time window as the analysis period;

[0088] The first and second eigenma matrices are decomposed using spectral decomposition, resulting in multiple eigenvectors and corresponding eigenvalues. These eigenvectors are labeled as the first group and the second group based on their respective eigenma matrices.

[0089] We introduce cosine distance to calculate the similarity between two sets of feature vectors. Specifically, we select a feature vector from the first set of feature vectors and calculate the average distance D between it and the second set of feature vectors.

[0090] Calculate the average distance D between all feature vectors in the first group and the second group, and average all D values ​​as the difference between the two groups of vectors.

[0091] The time window is moved cyclically, with a step size of N1 time periods, and the difference between the two sets of vectors corresponding to each time window is calculated.

[0092] The time windows with differences exceeding a preset threshold are marked, and the corresponding time periods are designated as noise time periods.

[0093] Here, it can be understood that cosine distance is used to represent the difference and similarity of feature vectors. The average distance D is calculated as follows:

[0094] ;

[0095] Where n is the number of vectors in the second set of feature vectors. This represents a selected feature vector, and DT() represents the calculation of cosine distance. Let represent the i-th vector in the second set of feature vectors, and D be the average distance.

[0096] N1 is the step size set by the user.

[0097] The difference value here can be set by the user selecting the vector cosine distance of the first and second fish groups' features at a certain time period as a benchmark. Setting a higher value means that the edge parameters are more dependent on sonar data, thereby blurring the edge areas for image segmentation correction.

[0098] It can be understood that the larger the difference value, the lower the similarity. Furthermore, it indicates that there are differences between the fish school features identified by sonar and the fish school features identified by image recognition in multiple time periods. Furthermore, by dynamically adjusting the edge segmentation parameters, the image recognition and segmentation situation can be adjusted, and dynamic adjustment of image recognition can be achieved.

[0099] According to an embodiment of the present invention, S4 specifically includes:

[0100] Based on the noise time period, a set of noisy images is selected from the image set;

[0101] The edge segmentation parameters are dynamically set according to the difference value corresponding to the time window, and secondary fish recognition is performed based on the preset image recognition model. The recognition results are updated in real time, and the corrected fish recognition results are sent to the preset terminal.

[0102] This means that the difference values ​​can be divided into intervals. For example, multiple intervals of difference values ​​can be pre-defined, each representing a different convolutional kernel and gradient magnitude threshold parameter. The corresponding edge segmentation parameters are then dynamically set based on the actual difference values. Higher difference values ​​correspond to larger convolutional kernel sizes and smaller gradient magnitude thresholds, dynamically adjusted based on the magnitude of the difference values. For example, the kernel size can be increased from 3×3 to 5×5.

[0103] In this dynamic parameter adjustment, increasing the convolution kernel improves edge detection capability and expands the range of identifiable fish images, but it will reduce edge localization accuracy to some extent. In addition, lowering the threshold introduces some weak edges to increase the identifiable area and improve recognition accuracy, but it will introduce noise to some extent.

[0104] It is worth mentioning that in the traditional underwater image recognition process, there are often large recognition errors in image segmentation, edge detection, and classification of target economic fish. This is generally due to the complex underwater environment, fluctuations in light intensity, and the presence of obstacles, which makes the image noise points complex and difficult to predict. Furthermore, existing technologies lack corresponding dynamic adjustment methods for edge detection, resulting in weak image segmentation capabilities and difficulty in controlling recognition accuracy. On the other hand, using sonar alone for fish school detection can often only detect data such as fish density and range, and cannot efficiently and quickly detect and identify fish species.

[0105] In addition, underwater image recognition methods can intuitively obtain detailed information such as the species and shape of fish by collecting underwater fish images and performing intelligent analysis. However, the edge segmentation parameters in traditional image recognition technology are mostly fixed values, which cannot adapt to the dynamic changes in underwater light intensity and water transparency. This results in incomplete segmentation of fish images and serious background noise interference, which in turn affects the accuracy of fish feature assessment.

[0106] Based on this, the present invention uses sonar and camera devices to perform multi-source data analysis on the target water layer. Fish school characteristics are evaluated through both sonar analysis and image analysis. Data is collected and matrixed based on fish school characteristics across multiple time periods. The differences between the two dimensions of fish school characteristics are evaluated using the feature matrix across multiple time periods. Furthermore, time windows are fused to perform resource assessment difference analysis at multiple consecutive time points, and time periods with significant noise and low recognition accuracy are selected. Further, the segmentation parameters are dynamically adjusted using the differences between the two dimensions of fish school characteristics to effectively control the accuracy of economic fish image recognition, avoid large fluctuations in recognition result accuracy, and reduce errors in resource recognition across multiple time periods.

[0107] Here, the noisy time period represents a certain deviation between the acoustic and image recognition characteristics of fish schools. This deviation is formed by multiple time points. During this period, adjusting the segmentation parameters and conducting secondary recognition evaluation on the corresponding image set can dynamically adjust the recognition results and make the classification and statistical results of economic fish in the images more accurate.

[0108] In addition, this invention dynamically adjusts edge segmentation parameters to adapt to the image set recognition needs of different underwater environments and complex environments, and improves the accuracy of fish school image recognition by combining acoustic fish school assessment.

[0109] According to an embodiment of the present invention, the preset terminal includes a computing terminal and a mobile terminal.

[0110] According to an embodiment of the present invention, both the preset monitoring period and the time window include multiple time periods.

[0111] Figure 3 The diagram shows a block diagram of a marine economic fish resource assessment platform based on multi-source data fusion according to the present invention.

[0112] like Figure 3 As shown, based on functional block division, the platform includes: a data acquisition module 101, an image processing module 102, a sonar processing module 103, and a preset terminal 104. The data acquisition module is used to collect sonar and image sets and generate relevant feature data; the image processing module is used to identify and classify image sets and dynamically adjust edge segmentation parameters; the sonar processing module is used to analyze and process sonar data and generate relevant feature data; the preset terminal is used to visualize the identification results and resource assessment results of economic fish species.

[0113] A second aspect of the present invention also provides a marine economic fish resource assessment platform based on multi-source data fusion. The platform includes a memory and a processor. The memory includes a marine economic fish resource assessment program based on multi-source data fusion. When executed by the processor, the marine economic fish resource assessment program based on multi-source data fusion performs the following steps:

[0114] S1: Within the target sea area, set the target water layer and collect data within a preset monitoring period. Collect sonar detection data based on the sonar detection device, and evaluate the fish school characteristics of economic fish based on the sound wave characteristic feedback to generate the first fish school characteristics for multiple time periods.

[0115] S2: The underwater camera device is used to collect images of the target water layer to obtain an image set. Edge segmentation parameters are set in the preset image recognition model. The image set is segmented, identified and evaluated for various economic fish species, and fish school characteristics are generated for multiple time periods.

[0116] S3: Set time windows based on preset monitoring cycles, combine the characteristics of the first fish group and the characteristics of the second fish group to generate the first feature matrix and the second feature matrix for each time window, use spectral decomposition to determine the similarity of the feature matrices within each time window, and filter out the noise time periods based on the similarity.

[0117] S4: Select a set of noisy images from the image set based on the noise time period, and dynamically adjust the edge segmentation parameters according to the similarity to perform secondary fish identification, and dynamically correct the fish identification results.

[0118] When the system / platform is running, it can perform one or more steps of the above-described method for assessing marine economic fish resources based on multi-source data fusion.

[0119] A third aspect of the present invention also provides a computer-readable storage medium comprising a marine economic fish resource assessment program based on multi-source data fusion, wherein when the marine economic fish resource assessment program based on multi-source data fusion is executed by a processor, it implements the steps of the marine economic fish resource assessment method based on multi-source data fusion as described in any of the preceding claims.

[0120] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this application can be generated, in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic cable, data subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital universal optical disc), or a semiconductor medium (e.g., solid-state drive). In the various embodiments of this application, unless otherwise specified or logically conflicting, the terminology and / or descriptions between different embodiments are consistent and can be mutually referenced. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0121] In the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone, where A and B can be singular or plural. In the textual description of the embodiments of this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship. In this application, "first," "second," and various numerical designations are only for the convenience of description and are not used to limit the scope of the embodiments of this application. For example, they are used to distinguish different messages, rather than to describe a specific order or sequence.

[0122] It is understood that the various numerical designations used in the embodiments of this application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application. The order of the process numbers does not imply the order of execution; the execution order of each process should be determined by its function and internal logic.

[0123] Finally, it should be noted that the above description is only a specific implementation of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the protection scope of this application.

Claims

1. A method for assessing marine economic fish resources based on multi-source data fusion, characterized in that, include: S1: Within the target sea area, set the target water layer and collect data within a preset monitoring period. Collect sonar detection data based on the sonar detection device, and evaluate the fish school characteristics of economic fish based on the sound wave characteristic feedback to generate the first fish school characteristics for multiple time periods. S2: The underwater camera device is used to collect images of the target water layer to obtain an image set. Edge segmentation parameters are set in the preset image recognition model. The image set is segmented, identified and evaluated for various economic fish species, and fish school characteristics are generated for multiple time periods. S3: Set time windows based on preset monitoring cycles, combine the characteristics of the first fish group and the characteristics of the second fish group to generate the first feature matrix and the second feature matrix for each time window, use spectral decomposition to determine the similarity of the feature matrices within each time window, and filter out the noise time periods based on the similarity. S4: Select a set of noisy images from the image set based on the noise time period, and dynamically adjust the edge segmentation parameters according to the similarity to perform secondary fish identification and dynamically correct the fish school identification results. Specifically, S3 refers to: A time window is set based on the preset monitoring cycle and the preset interval number N; A time window includes multiple time periods; Within a time window, the first feature matrix under that time window is constructed using time points as one-dimensional information and the characteristics of the first fish group as two-dimensional information. Construct the second feature matrix for this time window based on the features of the second fish swarm; S3 further includes: Introduce a time window as the analysis period; The first and second eigenma matrices are decomposed using spectral decomposition, resulting in multiple eigenvectors and corresponding eigenvalues. These eigenvectors are labeled as the first group and the second group based on their respective eigenma matrices. We introduce cosine distance to calculate the similarity between two sets of feature vectors. Specifically, we select a feature vector from the first set of feature vectors and calculate the average distance D between it and the second set of feature vectors. Calculate the average distance D between all feature vectors in the first group and the feature vectors in the second group, average all D values ​​and use them as the difference between the two groups of feature vectors; The time window is moved cyclically, with a step size of N1 time periods, and the difference between the two sets of vectors corresponding to each time window is calculated. The time windows with differences exceeding a preset threshold are marked, and the corresponding time periods are designated as noise time periods.

2. The method for assessing marine economic fish resources based on multi-source data fusion according to claim 1, characterized in that, Specifically, S1 is: Monitoring devices are set up according to the target sea area, and water monitoring is carried out on the target water layer within a preset monitoring period; Based on the sonar detection data collected by the sonar detection device, the sonar detection data is processed by adaptive median filtering to obtain time-domain sonar data; The time-domain sonar data is converted into sonar feature information, including frequency domain data, by Fourier transform, and the amplitude, frequency, phase and sonar intensity features of the sound wave are extracted. By combining the fish sonar feature database, the sonar feature information is used to identify economically important fish species and evaluate fish characteristics to obtain information on fish school size, fish school density, and fish school quantity. Based on the monitoring frequency, the preset monitoring period is divided into multiple time periods. For each time period, the corresponding fish school range size, fish school density, and fish school number information are analyzed as the first fish school feature in multiple dimensions.

3. The method for assessing marine economic fish resources based on multi-source data fusion according to claim 1, characterized in that, Specifically, S2 is: Based on an underwater camera device, images are acquired at a preset monitoring cycle for the target water layer, and an image set is obtained; The image acquisition frequency is kept consistent with the sonar detection and analysis frequency; The image set is preprocessed by grayscale conversion, enhancement, and noise reduction; In the preset image recognition model, the Sobel edge detection operator is used to perform image edge detection and image segmentation, and the default edge segmentation parameters are set. Edge segmentation parameters include the kernel size and gradient magnitude threshold; The Sobel edge detection operator is used to detect edges in the image set and define edge points to form edge regions. Fish targets are segmented and classified based on edge regions to obtain fish identification results; Based on the fish identification results, the fish school feature information of multiple time periods is statistically analyzed in the image, and a second fish school feature is generated.

4. The method for assessing marine economic fish resources based on multi-source data fusion according to claim 1, characterized in that, Specifically, S4 is: Based on the noise time period, a set of noisy images is selected from the image set; The edge segmentation parameters are dynamically set according to the difference value corresponding to the time window, and secondary fish recognition is performed based on the preset image recognition model. The recognition results are updated in real time, and the corrected fish recognition results are sent to the preset terminal.

5. The method for assessing marine economic fish resources based on multi-source data fusion according to claim 4, characterized in that, The preset terminals include computing terminals and mobile terminals.

6. The method for assessing marine economic fish resources based on multi-source data fusion according to claim 1, characterized in that, The preset monitoring period and time window both include multiple time periods.

7. A marine economic fish resource assessment platform based on multi-source data fusion, characterized in that, The platform includes a memory and a processor. The memory includes a marine economic fish resource assessment program based on multi-source data fusion. When the processor executes the marine economic fish resource assessment program based on multi-source data fusion, it implements the steps of the marine economic fish resource assessment method based on multi-source data fusion as described in claim 1.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a marine economic fish resource assessment program based on multi-source data fusion. When the marine economic fish resource assessment program based on multi-source data fusion is executed by a processor, it implements the steps of the marine economic fish resource assessment method based on multi-source data fusion as described in any one of claims 1 to 6.

Images